# A CNN-based methodology for breast cancer diagnosis using thermal images

**Authors:** Juan Zuluaga-Gomez, Zeina Al Masry, Khaled Benaggoune, Safa Meraghni, Noureddine Zerhouni  
**arXiv:** [1910.13757](https://arxiv.org/abs/1910.13757)  
**Date:** 2019-10-30  
**Summary page:** https://www.summarizepaper.com/en/paper/1910.13757v1/

## Key points

- Study focuses on developing a computer-aided diagnosis system for breast cancer using CNN with thermal images
- Motivation: Global burden of breast cancer, with two million women diagnosed in 2018
- Aim to demonstrate that CNN-based CAD systems offer faster, more reliable, and robust alternative for diagnosis
- Experiments evaluated impact of data pre-processing techniques and database size on CNN model performance
- Developed optimization algorithm for fine-tuning CNN hyperparameters using tree parzen estimator
- Results showed high accuracy (92%) and F1-score (92%), outperforming established architectures like ResNet50, SeResNet50, and Inception
- Data augmentation techniques enabled comparable performance metrics even with smaller database size
- Emphasizes advantages of utilizing data augmentation and CNNs in analyzing thermal breast images for diagnosing breast cancer
- Importance of considering database size when evaluating performance of CNN models in medical imaging applications

## Summary

The study conducted by Juan Zuluaga-Gomez, Zeina Al Masry, Khaled Benaggoune, Safa Meraghni, and Noureddine Zerhouni focuses on developing a computer-aided diagnosis system for breast cancer using convolutional neural networks (CNN) with thermal images. The motivation behind this research is the global burden of breast cancer, with two million women diagnosed in 2018 according to GLOBOCAN data. Traditional screening methods like mammography, MRI, ultrasound, and blood tests are often inaccessible due to cost or lack of qualified personnel in certain regions. The researchers aimed to demonstrate that CNN-based CAD systems offer a faster, more reliable and robust alternative for breast cancer diagnosis. They conducted experiments to evaluate the impact of data pre-processing techniques and database size on the performance of CNN models. Additionally, they developed an optimization algorithm for fine-tuning CNN hyperparameters using a tree parzen estimator. Results from their study involving 57 patient cases showed that their CNN models achieved high accuracy (92%) and F1-score (92%), outperforming established architectures like ResNet50, SeResNet50 and Inception. Furthermore, they found that implementing data augmentation techniques enabled comparable performance metrics even with a smaller database size. In conclusion,the study highlights the advantages of utilizing data augmentation and CNNs in analyzing thermal breast images for diagnosing breast cancer. It emphasizes the importance of considering database size when evaluating the performance of CNN models in medical imaging applications. The findings contribute to advancing diagnostic methodologies for breast cancer detection using innovative technologies like convolutional neural networks.